Environmental Evidence Synthesis

We make a number of decisions every day. These decisions are based on conscious or unconscious analysis, as well as assessments of risks and uncertainties. We are often unaware of how we make these decisions, and how we compare alternatives. Individuals, as well as society, sometimes make decisions to be on the safe side, whereas in other cases chose to take a risk. The course focuses on a number of issues: How can risk be assessed? Can risk be quantified? How do risk assessments affect decision and policy making? What role does risk communication and media play? How do individuals perceive risks and risk management? Risk, uncertainty, and decision analysis implies systematic efforts to understand the consequences of decisions.

The aim of the course is to give the participants a deeper understanding of theoretical perspectives and methods within risk research in different disciplines, as well as tools for interdisciplinary risk research. The course contains both lectures and seminars. The lectures will provide rich accounts of different disciplinary perspectives. In seminars, participants will cooperate around certain issues, which demand an interdisciplinary approach.

Lectures by Ragnar Löfstedt, Annika Wallin, Åsa Knaggård, Joakim Zander, Johannes Persson, Henrik Thorén, Maj Rundlöf and Ullrika Sahlin

Physical meeting in Lund 2-3rd October 2023

Hybrid meeting 6-7 November 

Read more and apply here before Sept 18th 2023 course page 

Den 17 maj kommer Ullrika medverka i en panel på detta tema på Formas Agenda2030 konferens i Stockholm.

Are you interested in modelling biodiversity to inform decision-making?

Would you like to learn more about machine learning and Bayesian statistical modelling?

Then this PhD position could be interesting for you! Apply here before May 5th 2023!

This PhD project is part of the programme BIOPATH “Pathways for an efficient alignment of the FInancial System with the needs of Biodiversity” , funded by the Foundation for Environmental Strategic Research. BIOPATH aims to assess, develop and test existing and novel approaches for the integration of biodiversity into financial decision-making in collaboration with financial and industrial partners. The PhD student will be involved with developing and using quantitative models to assess trends and impacts on biodiversity resulting from human activities including efforts to conserve or restore biodiversity. The modelling to be done by the PhD student encompasses artificial intelligence as well as process-based statistical methods to assimilate biodiversity data in the context of causal analysis and optimisation of conservation action.

The PhD student is expected to review and develop methodological aspects, to consider alternative biodiversity metrics, to handle knowledge gaps and to characterise uncertainty. The performance of the models developed in this project in a real-world scenario will be evaluated in a case study about integrating biodiversity in forest management.

Part of the project is devoted to apply CAPTAIN – Conservation Area Prioritization Through Artificial Intelligence – to quantify trade-off between costs and benefits of area and biodiversity protection, using multiple biodiversity metrics.

Another part is to work with joint species distribution modelling to build statistical models to evaluate impacts on biodiversity.

This project will be supervised by Ullrika Sahlin (Lund University), Daniele Silvestro (University of Fribourg and University of Gothenburg), and Henrik Smith (LU) but will involve strong interactions with other leading scientists within the BIOPATH program, in Sweden and abroad. The work may include research visits to the University of Gothenburg, University of Fribourg (Switzerland), and University of Exeter (UK).

Dear fellow Bayesians!

Registration for the upcoming Bayes@Lund 2023 is now open! The purpose of Bayes@Lund is to bring together researchers and professionals working with or interested in Bayesian methods. The conference aims at being accessible to researchers with little experience in Bayesian methods while still being relevant to experienced practitioners.

Date: Monday, 23 January 2023, 10:00 – 17:00 CET
Place: Blå Hallen, Ekologihuset, Lund University
Web: www.bayesatlund.com

Thanks to everyone who contributed their abstracts, we are now looking forward to an exciting event that covers many topics from the philosophy of probability to Bayesian methods and applications. The program consists of a rich mix of keynotes, tutorials, and contributed presentations! Please, check out the final conference program and register your attendance at the Bayes@Lund 2023 website. The conference is free to attend.  We are looking forward to seeing you there soon!

All best,

The committee Ullrika Sahlin, Dmytro Perepolkin and Rasmus Bååth

We are approaching the final set of training courses on Expert Knowledge Elicitation offered during 2022-2023

In June 2014, EFSA published a Guidance document on Expert Knowledge Elicitation (EKE) in Food and Feed Safety Risk Assessment. In this context, EKE is defined as a systematic, documented and reviewable process to retrieve expert judgments from groups of experts in the form of probability distributions.

EKE methods are formal, probabilistic judgment techniques designed to encourage careful, thoughtful judgments and reduce psychological biases. EFSA Guidance implements EKE in an efficient, rigorous and transparent manner, targeted on most important parameters, subject to critical review at key decision points, and fully documented.

Lund University currently offer online training courses on Expert Knowledge Elicitation (EKE)

There will be four courses offered during spring 2023 – more info

We are happy to announce the mid-term seminar for Dmytro Perepolkin, PhD student, at CEC, Lund University on October 19th 2022 at 9.15 CET

At this seminar – Dmytro will present work he has done and will be doing in his PhD thesis, which will be followed by a discussion with a mid-term opponent.

The mid-term opponent is Professor Andrew Robinson at the Centre of Excellence for Biosecurity Risk Analysis (CEBRA), and a Professor in applied statistics at the University of Melbourne.

Recording of Dmytro’s excellent presentation 

Abstract

Expert Knowledge Elicitation is an emerging field in applied sciences particularly useful in situations when the data is sparse or the system under investigation is complex, as is the case with environmental management problems. We propose some new probability distributions for making the elicitation of expert judgment easier, as well as to simplify their integration into common statistical models for assessment supporting decision-making. We show an application of a model parameterized by the elicited quantiles to food safety and outline the path towards using the quantile-parameterized distributions in complementing the presence-only data for species distribution modelling.

Keywords: quantile-based Bayesian inference, quantile-parameterized distributions, exposure assessment, species distribution models, GBIF

One year to go before we open up the 2023 conference for the Society for Risk Analysis Europe in Lund.

Dates 18 to 21 June 2023

The theme of the conference is Risk and assessment in a changing world

Funding from The Crafoord Foundation is highly appreciated.

I have a poster at SETAC Copenhagen about Bayesian networks. My main message is that Bayesian networks should not be limited to specific applications of probabilistic graphical models, and that the term is currently confusing including both expert informed Bayesian Belief networks and data-driven Bayesian networks in machine learning. I do not see any reason why not Bayesian statistical models can not be seen as a Bayesian network. Widening the concept can benefit from the increasing uptake and positive response of Bayesian networks to support the need for flexible quantitative models in science and society. With this in mind, there are different modelling approaches able to do different things that are Bayesian networks. We then have to go back and describe what each approach actually does. What do I gain from this? Well. at least I am better off when we include any probabilistic graphical modelling into the concept of Bayesian network.

Check out my poster and talk part of the survey giving your perspective on what is (or is not) a Bayesian network.

Should We Reconsider or Upgrade Bayesian Networks for Environmental Assessments_poster_SETAC2022

The survey – link to google forms

Results can be seen here

Ivette’s work A robust Bayesian bias-adjusted random effects model for consideration of uncertainty about bias terms in evidence synthesis has been accepted for publication in Statistics in Medicine.

DOI: 10.1002/sim.9422

In this paper, we construct a set for the bias terms based on the qualitative information in a Cochrane risk of bias table. To accommodate the imprecision in bias terms, bias adjustment in a meta-analysis model is performed in a Robust Bayesian analysis framework which results in lower and upper bounds on the probability expressing uncertainty in quantities of interest.

Congratulations

Following the succes from 2021 we will continue with series of online talks as part of Bayes@Lund

Welcome to join

Organisers

Dmytro Perepolkin, Ullrika Sahlin, Rasums Bååth

Aubrey Clayton  on Bernoulli’s Fallacy

February 3rd, 2022 at 16 CET on Zoom.

Abstract:
For as long as there has been statistical inference, practitioners have been confusing the probability of data-given-hypothesis and the probability of hypothesis-given-data, the “fallacy of the transposed conditional.” In this talk, I’ll survey the history of this mistake including recent examples in disparate fields, up to and including present-day orthodox statistical practice. I’ll argue that the fallacy represents a larger failure of thinking about probability–which I call “Bernoulli’s Fallacy”–and that the same error underlies much of the current crisis of replication in science. Finally, I’ll share thoughts about what we should do to ensure the statistical methods of the future are built on a solid foundation.